I’m trying to improve on my ML-based bird recognition results. Currently, a Raspberry Pi with a camera detects motion at the feeder, and sends pictures to the AWS Rekognition service, which I’m using to try to identify the species of bird. But I’d like to improve on the accuracy of AWS Rekognition, while keeping the cost down. So here, I’m using fastai, which is a high-level Python library, which runs on top of Pytorch. So far, I’m getting over 80% accuracy on the 200-bird Caltech-UCSD bird image dataset. You can check out my model on Kaggle, here:
Note that (as I describe in the next post), a big issue here is that I used transfer learning, starting with the Resnet model. Apparently Resnet’s already seen some of these images, so my 80%+ accuracy is probably an overestimate.